In a remarkable example of tech companies both competing and cooperating, OpenAI will begin to rent compute infrastructure from Google Cloud. Reflecting the agreement’s unusual nature, neither company has yet formally announced the arrangement despite widespread press coverage.
At one level the deal is routine, even expected. OpenAI, as the developer of ChatGPT—with its voracious and ever growing requirements for compute—needs to continuously find more sources of AI infrastructure. The only possible sources of compute power at this level are the cloud hyperscalers, a very short list. Given that OpenAI is already hosted on Microsoft Azure and Oracle Cloud but still needs more capacity, Google Cloud is the obvious choice.
But there’s a surprising element to the OpenAI-Google deal due to the multiple levels of competition between the two AI leaders. Both companies offer an AI chatbot, and for both, this product is central to their business model.
Google has poured resources into its Gemini AI chatbot, but it faces tough competition from OpenAI’s ChatGPT, widely seen as the AI market’s leader. OpenAI recently announced that its annualized revenue run rate has hit an impressive $10 billion.
Gemini remains in the running, though determining accurate user numbers for AI chatbots is a difficult because the numbers are self reported. Brad Lightcap, OpenAI’s Chief Operating Officer, said in March that ChatGPT has 400 million weekly users, with plans to hit 1 billion by the end of 2025. Google CEO Sundar Pichai said in late May that Gemini has more than 400 million monthly active users, so presumably it has fewer weekly users than ChatGPT.
On the other hand, at ChatGPT’s debut in late 2022 and for months later it had no competitors, yet now Gemini has a large user base, so it’s likely pulling users away from ChatGPT.
In any case, the competition between these two companies goes deeper than market share of chatbot users. Google earns roughly 75% of its revenue from advertising, the majority of which comes from search. OpenAI poses an existential threat to Google’s search business because ChatGPT offers a simpler, streamlined and—in the view of many users—more informative replacement for Google search.
Google is working to combat this encroachment into its search business, particularly by implementing “AI Overviews” that use AI in an attempt to provide a fully formed answer to users’ search query. Whether this effort will succeed long term is an open question.
In the meantime, the Google-OpenAI deal is a plus for Google Cloud, even it may not be for Alphabet itself long term. Early reports suggest that Google will support both training and inference for OpenAI’s models.
This deal by OpenAI appears to be part of a larger effort to broaden out its sources of compute infrastructure. It formed a partnership with CoreWeave for AI infrastructure in March for $11.9 billion and also become an investor in the company. OpenAI gets substantial hosting from Oracle Cloud as part of the high profile, $500 billion Stargate initiative launched in January.
Providing OpenAI with even more compute capacity will be a separate Stargate UAE project, which is a planned massive data center facility in Abu Dhabi that is expected to host 5 gigawatts of capacity. Partners include OpenAI, Nvidia, Oracle, Cisco and Softbank.
OpenAI continues to work with Microsoft, its original infrastructure partner with which it launched ChatGPT in November 2022. Reports suggest that Microsoft and OpenAI—which also compete with each other in the AI market—are renegotiating the terms of their partnership.
